Replace with per-group INT8 quantization (group_size=128, lossless vs FP16)

0140a80 2 months ago

187 Bytes

	{
	"base_model": "apple/FastVLM-1.5B",
	"quantization": "pytorch_int8",
	"bits": 8,
	"method": "per-group symmetric",
	"group_size": 128,
	"vision_encoder": "fp16 (not quantized)"
	}